Download
Software
All the necessary software tools are available as a single package, pyClarity available on GitHub.
Data
To download data, please visit here.
The data is split into two packages:
clarity_CEC1_data.train.tgz
[192 GB],clarity_CEC1_data.dev_eval_metadata.tgz
[163 GB].
Please also download and unpack clarity_CEC1_data.anechoic.v1_3.tgz
[11.4 GB], which contains the correct version of anechoic signals for reference, and use them to replace the incorrected anechoic signals within the clarity_CEC1_data.train.tgz
and clarity_CEC1_data.dev_eval_metadata.tgz
packages.
Unpack packages under the same root directory using
tar -xvzf <PACKAGE_NAME>
Training data is stored in clarity_CEC1_data.train.tgz
with the following structure,
clarity_data
|
└───train
└───interferers
| | nosie 3.9G
| | speech 4.5G
|
└ ───rooms
| | ac 48M
| | brir 46G
| | rpf 379M
|
└───scenes 166G
|
└───targets 2.8G
The dev and eval data is stored in clarity_CEC1_data.dev_eval_metadata.tgz
contains
clarity_data
|
└───dev
| └───interferers
| | | nosie 587M
| | | speech 1.4G
| |
| └───rooms
| | | ac 20M
| | | brir 20G
| | | rpf 158M
| |
| └───scenes 72G
| |
| └───targets 1.3G
|
└───eval
| | | nosie 675M
| | | speech 1.3G
| |
| └───rooms
| | | ac 12M
| | | brir 12G
| | | rpf 95M
| |
| └───scenes 58G
| |
| └───targets 749M
|
└───eval2/scenes 21G # eval signal processed by baseline system
For further details and code for running the baseline, see the baseline recipe in the pyClarity repository.